Abstract Background Transoral robotic surgery (TORS) is increasingly used for head and neck pathologies, including lateral oropharyngectomy for tonsillar carcinomas. With expanding use across robotic platforms (Da Vinci, Versius), there is growing potential to apply Artificial Intelligence (AI) and Surgical Data Science to analyse operative video. By examining workflow at multiple levels, from broad surgical phases to granular gestures, AI has the potential to support intraoperative decision-making, enhance training, and improve patient outcomes. Methods Eight TORS lateral oropharyngectomy cases at Guy’s Hospital (London, UK) were recorded via the Proximie platform. A 7-step protocol was developed to annotate surgical phases, and gestures were manually labelled with VIA software to quantify operating time, energy device use, and workflow. An automated pipeline was built to extract video frames across surgical gestures and curate datasets for AI model training. Results Median operation time was 87 min (mean 102 ± 40). The most time-consuming stages were posterior extension (mean 32.2 ± 18.8 min) and tongue base resection (mean 19.2 ± 18.1 min). Across cases, 3261 gestures were annotated; the most frequent was ‘hot cut’ (thermal dissection, n = 1600). Active operating time comprised 41.2 ± 8.2% of total duration. A scalable pipeline was established to enable automated AI-based recognition of surgical gestures. Conclusions This is the first study to apply Surgical Data Science to workflow analysis in TORS. It provides a foundation for integrating AI into head and neck interventions, with potential to enhance training, deliver data-driven operative feedback, and assess surgical performance in relation to patient outcomes.
Kasetti et al. (2026) studied this question.